Note: I wrote this column with my husband, Josh.
In May of 1997, the computer Deep Blue beat the world chess champion. In February of 2011, the computer Watson beat the world Jeopardy champions. Someday in the future, humans such as you and I will probably not be the smartest beings on Earth.
Martin Hilbert and Priscila López, in a 2011 paper in the journal Science, estimate that the total computing ability of the world’s computers passed the computational ability of a single human brain in 2007. They also estimate that the combined computing power of humanity’s general-purpose computers have been growing at a compound annual growth rate of 58% between 1986 and 2007. While it is difficult to compare human brains with computer computation with precision, the Hilbert and López calculation, if anything, appears to be over-estimating the computing power of a single brain, so the combined power of human computers very well may have passed a single human earlier than 2007. In short, humanity could very well already create a computing system that was smarter than any human if we just connect enough existing computers and knew how to program it.
Every day, there are computer scientists working to create computers that search the Internet better, that can identify photos and people in those photos, that can create reports for humans from knowledge found on the Internet, cars that tell their owners when they need maintenance, and so forth. Every computer that is created to be better at interacting with humans and bringing humans the knowledge they search for brings closer the day of truly intelligent computers and robots that are smarter than humans.
We believe two things are almost inevitable: Intelligent computers and that they will not obey us. We have no reliable way to keep intelligent robots from turning on us ― we can’t make something more intelligent than humans our slaves (at least not for long). And if you think they can be programmed not to hurt us, think about how many bugs are in the average computer program, which is much less complicated. Most people think in terms of Utopia when thinking of robots ― we can make them do what we want when we want ― in reality they will be much too powerful. The more we expect from our computers the closer we bring the day of reckoning ― smarter than human artificial intelligence. Is this a good thing?
How it turns out for humanity depends on how the artificial intelligence treats us. This relationship could range from the robots being helpful when they wish to the robots attempting to destroy us. We don’t know how it will all play out but we think this needs to be discussed more.
The above essay was originally published in the (Idaho Falls) Post Register on June 16, 2011.
Additional Comments by Josh:
Human brains seem to be made up of components that are much bigger than comparable computer components. Human brains are much more efficient energy wise. Therefore, all that may be required for smarter than human robots is an increase in energy efficiency, without any more improvements in computing speed or size. Note also that since electric signals can move at the speed of light (3e8 m/s), 1.5 million times faster than nerve impulses, so a computer could be over 100 miles long and still have communication across it as fast as a human brain.
The belief that humans can somehow contain computers, either by limiting their actions through programming or by limiting their access to physical control, is probably incorrect. Think about how many loopholes are in things like human laws, and remember that lawyers are only human. Limiting the robots (such as by providing them only with a computer monitor for output) would only last until the computer managed to either trick a human into doing something that seems innocuous, or they manage to do something else tricky (such as perhaps using tempest radiation for radio transmission and receiving).
The way I think about a intelligent computer, is that it could do the kinds of thinking that a human could do, just faster. So imagine you could ask a hundred people a question. They think about it, and then come up with an answer. So a computer with a hundred times the computational ability of a human would be able to think as fast as a hundred people, except it would probably be even faster since it could coordinate a response better. So a question that would take an hour to answer would be answered in about half a minute.
Humans do take care of less intelligent animals, but we call them pets.
It might be possible to avoid having smarter than human computers (if everyone were Amish, this would not be a problem), but this would require serious restrictions on technology. Basically, this would require restricting the total computational power of humanity. If the computational power is not restricted, then it becomes possible that someone could accidentally create an intelligent computer.
An intelligent malevolent computer connected to the internet could do serious damage. Most of humanity’s industrial capacity is connected to the internet in some way. Almost all of humanity’s telecommunications capacity is connected to the internet.
One key question is how soon human level intelligent computers appear. If Hilbert and López are correct that humanity had the computational power of a human in 2007, and this continues to grow at 58% a year, then every 5 years, the computational power grows by about a factor of 10 (1.58^5 = 9.8). So in 2012, the world computational power is 10 human brains, in 2017 it is 100 times, in 2022 it is 1000 times, in 2027 it is 10,000 times. Diverting 1/1000 of human’s computers is much easier than diverting 100% of the the computers.
Nerve speeds: ~100 m/s ~200 m/h
The Control of Neuron Number. Robert W. Williams and Karl Herrup, 2001,
300 neurons – nematode worms ( Caenorhabditis elegans ) http://wormweb.org/neuralnet
Human brain – at least 10 billion neurons, perhaps as many as 1 trillion, most likely 95–100 billion neurons
Complexity of neuturons: Probably not that great. Just three differential equations in the Hindmarsh-Rose model.
Size of Neurons:
Soma: 4 to 100 micrometers (http://en.wikipedia.org/wiki/Neurons and http://faculty.washington.edu/chudler/facts.html#neuron )
Axon and dendrites: 1 micrometer thick.
Size of computer chip components (feature size):
45 nanometers (0.045 micrometers) (http://en.wikipedia.org/wiki/Moore’s_law)
Computer processing power and growth rate:
6.4e18 instructions per second in 2007
Comparable to 10e17 nerve impulses in one human brain per second.
Growing 58% compounding annual growth rate since 1986
The World’s Technological Capacity to Store, Communicate, and Compute Information
Martin Hilbert and Priscila López
Science 332, 60 (2011);